A clash resolved in a federated model costs a few hours of engineering time. The same clash resolved on site costs materials, labour, programme and often a variation claim. Everyone knows this, and projects still discover the majority of their conflicts during construction.
Why it keeps happening
Late coordination is rarely a modelling failure. It is a sequencing failure — disciplines are appointed at different stages, work to different milestones, and exchange information at contractual rather than practical intervals.
The structural fix
Appointing disciplines together and running one federated model from concept removes the exchange problem entirely. Where that is not possible, defining coordination gates tied to design maturity rather than calendar dates is the next best control.
Measuring it
Track clash counts by stage. A project resolving most conflicts before tender is behaving well; a project whose clash count peaks during construction has a process problem, not a modelling one.